Timely identification of melanoma is vital for enhancing patient prognosis, nevertheless, it continues to be a formidable undertaking. This study presents a hybrid DL model that combines attention mechanisms, LSTM networks, and CNNs to increase the precision and effectiveness of melanoma classification. To facilitate instantaneous photo collecting and analysis, the model is integrated into a smartphone application. For training and testing, the ISIC Archive dataset—which contains over 25,000 dermoscopic images—was used. With an accuracy rate of 95.2%, precision rate of 94.8%, recall rate of 95.5%, and F1-score, the model showed notable gains in performance metrics of 95.1%. These findings exhibit a significant improvement compared to the current cutting-edge techniques, which showed poorer performance metrics in all areas. The significant increase in precision and other crucial measurements underscores the efficacy of the hybrid model and its prospective use in clinical environments for the early diagnosis of melanoma. Subsequent efforts will concentrate on including a wider range of datasets and enhancing the model’s efficiency for instantaneous processing on mobile devices.